GEOPHYSICAL RESEARCH LETTERS, VOL. 31, L10106, doi:10.1029/2004GL019539, 2004
Evaluation of tropical cirrus cloud properties derived from ECMWF
model output and ground based measurements over Nauru Island
Jennifer M. Comstock
Pacific Northwest National Laboratory, USA
Christian Jakob
Bureau of Meteorology Research Centre, Australia
Received 20 January 2004; revised 19 April 2004; accepted 28 April 2004; published 26 May 2004.
[1] Cirrus clouds play an important role both radiatively
and dynamically in the tropics. Understanding the
mechanisms responsible for the formation and persistence
of tropical cirrus is an important step in accurately
predicting cirrus in forecast and climate models. In this
study, we compare ground-based measurements of cloud
properties with those predicted by the ECMWF model at a
location in the tropical western Pacific. Our comparisons of
cloud height and optical depth over an 8 month time period
indicate that the model and measurements agree well. The
ECMWF model predicts cirrus anvils associated with deep
convection during convectively active periods, and also
isolated cirrus events that are influenced by large-scale
vertical ascent. We also show through examination of an
upper tropospheric cirrus case that the model produces
tropospheric waves that appear to influence the morphology
INDEX TERMS: 3374
and maintenance of the cirrus layer.
Meteorology and Atmospheric Dynamics: Tropical meteorology;
3360 Meteorology and Atmospheric Dynamics: Remote sensing;
3337 Meteorology and Atmospheric Dynamics: Numerical
modeling and data assimilation. Citation: Comstock, J. M., and
C. Jakob (2004), Evaluation of tropical cirrus cloud properties
derived from ECMWF model output and ground based
measurements over Nauru Island, Geophys. Res. Lett., 31,
L10106, doi:10.1029/2004GL019539.
1. Introduction
2. Clouds in the ECMWF Model
[2] Identifying the mechanisms responsible for the formation of cirrus clouds is important in understanding the
role of cirrus in the tropical atmosphere. Thin cirrus clouds
near the tropical tropopause transition layer (TTL) can have
significant impacts on the radiative heating in the upper
troposphere [McFarquhar et al., 2000]. These clouds may
also affect the transport of water vapor through the TTL into
the stratosphere [Rosenfield et al., 1998; Sherwood, 1999].
The ability of large-scale models to correctly simulate
tropical cirrus occurrence is important in predicting the role
of these clouds on climate. Climate models failure to
properly simulate cloudiness in the tropics has been identified as a critical improvement needed for the accurate
prediction of future climates [Wielicki et al., 2002]. In this
preliminary study, we explore the skill at which a largescale forecast model predicts the presence of tropical cirrus.
The motivation for this is twofold. First, assessing the
Copyright 2004 by the American Geophysical Union.
0094-8276/04/2004GL019539$05.00
model’s ability to simulate the observed characteristics of
tropical cirrus may identify important weaknesses in the
model. Second, if the model exhibits sufficient skill in
simulating these characteristics we might be able to use it
as a tool for identifying and understanding cirrus formation
mechanisms and large-scale features that are important in
maintaining extensive cirrus sheets that persist in the tropics
over long time periods. It is this understanding that is the
ultimate aim of our work. In this study we compare ground
based lidar and radar measurements of cirrus occurrence and
optical properties obtained at the Department of Energy’s
Atmospheric Radiation Measurement [ARM; Ackerman
and Stokes, 2003] site located on Nauru Island with those
generated by the European Centre for Medium Range
Weather Forecasts (ECMWF) model as a first step to assess
the accuracy of a forecast model in predicting tropical
cirrus.
[3] We first compare cloud height and visible optical
depth t derived from measurements and model simulations
over an 8-month period. Second, we discuss the different
cirrus types observed in the tropics and the dynamic
processes that are responsible for their formation. Finally,
we present a case study that illustrates the ability of the
model to predict thin tropical cirrus.
[4] For comparison with ground based measurements, we
use ECMWF model output, which has a 60 km horizontal
resolution. A time series of hourly model variables are
created using concatenated operational 12– 35 hr forecasts
for the grid point that includes the location of Nauru. The
cloud parameterization is a fully prognostic cloud scheme
[Tiedtke, 1993; Jakob, 2001] and includes a strong coupling
between the convection and cloud scheme. Both cloud
fraction and ice water content (IWC) have an explicit
convective source.
[5] Over the period of study, several changes were made
to the ECMWF model. First, the vertical resolution changed
from 50 levels to 60 levels on 13 October 1999, with extra
levels added primarily to the boundary layer. The ice cloud
parameterization was also changed in October 1999. The
initial cloud scheme had a single parameterization relating
ice crystal fall velocity (V) to IWC. The new cloud scheme
assumes the same V-IWC relationship for large particles
(>100 mm), but for small particles (<100 mm) a zero
fallspeed is assumed. These changes improved agreement
L10106
1 of 4
L10106
COMSTOCK AND JAKOB: TROPICAL CIRRUS CLOUD PROPERTIES
L10106
model output for a 60 km grid box with single point
measurements, although some uncertainty remains.
3. Cloud Statistics
Figure 1. Composite statistics comparing measured and
model cloud properties. (a) base height, (b) Cloud top
height, and (c) high cloud visible optical depth.
between the modeled and measured surface radiative fluxes
(not shown).
[6] The model produces a vertically varying profile of
cloud fraction, thereby generating a number of different
cloud situations within each model grid box. In order to
sample this sub-grid scale variability, each grid box is
subdivided into 100 sub-boxes to give 100 cloud configurations following an algorithm devised by Jakob and Klein
[1999]. Each of the 100 sub-boxes is treated as an independent sample of both the cloud geometry (cloud base and
top location) and optical thickness. Note that the IWC (or
liquid water content LWC) is assumed the same in each
cloudy sub-box for a given level so that differences in these
parameters arise solely from the cloud cover variations with
height by applying cloud overlap rules to each sub-box. The
threshold value of condensate for the existence of a cloud in
the model is 10 8 kg kg 1. The model calculates t using a
relationship between ice water path (IWP) and effective
radius [Ebert and Curry, 1992]. The latter is prescribed as a
function of temperature. Cloud boundaries are identified by
the presence of condensate (ice or water) in each atmospheric layer. The direct accounting for sub-grid scale
variability alleviates some of the problems of comparing
[7] In this study, we use the Active Remotely-Sensed
Cloud Locations (ARSCL) algorithm [Clothiaux et al.,
2000] to obtain cloud boundaries. This algorithm has
the benefit of combining both micropulse lidar [MPL;
Campbell et al., 2002] and millimeter cloud radar
[MMCR; Moran et al., 1998] observations to obtain
complete cloud boundaries. This is particularly important
in the tropics where high cirrus clouds near the tropical
tropopause are below the detection limit of the MMCR
because they are optically thin and likely consist of small
ice crystals [Comstock et al., 2002]. For cloud conditions
when low level boundary layer clouds lie below cirrus
layers, or optically thick anvil and middle level clouds, the
MMCR provides better estimates of cirrus cloud top height
because lidar is unable to penetrate through optically thick
layers. During 1999 at Nauru, low clouds block high
clouds approximately 11% of the time [Comstock et al.,
2002].
[8] ARSCL cloud boundary statistics between April and
November 1999 are compared with cloud boundaries output
from the ECMWF model. The cloud parameterization
modifications mentioned above improve the representation
of ice clouds in the ECMWF model; however, most of the
time period of this study lies before the change was made.
Therefore, we compensate for a model overestimation of
high cirrus cloud amount by removing cloud occurrences
when t < 0.005. This threshold was chosen by determining
that 95% of clouds with t < 0.01 produced by the ECMWF
model have t < 0.005. This removes residual small ice
contents that exist in small quantities in the model upper
troposphere.
[9] Comparisons between model and measured cloud
base frequency (Figure 1a) reveal very good agreement for
high and middle level clouds. Cloud base (zb) height
frequency for boundary layer clouds with zb < 2 km do
not agree as well. The majority of model cloud bases in
the boundary layer lie above 1 km, whereas measured
bases are typically below 1 km. Previous ECMWF model
and radar comparisons also found that the model tended to
have fewer clouds below 1 km and more clouds above
1 km as compared with measurements [Mace et al., 1998].
Cloud top height frequency of occurrence (Figure 1b) in
the upper troposphere does not agree as well as cloud base
frequency. Despite the removal of residual ice layers in
the upper troposphere from the results, the model has
significantly higher cloud tops than in the measured tops.
This implies that model clouds tend to be vertically
deeper.
[10] We use MPL measurements to obtain cloud visible
optical depth [Comstock and Sassen, 2001] for optically
thin cirrus clouds in the upper troposphere. Lidar measurements typically become attenuation limited when t
approaches 3.0. Previous studies have shown that radar
and lidar retrievals of optical depths compare well for this
time period at Nauru [Comstock et al., 2002]. The frequency of occurrence of t for model and measurements
(Figure 1c) is also in good agreement. The ECMWF
2 of 4
L10106
COMSTOCK AND JAKOB: TROPICAL CIRRUS CLOUD PROPERTIES
Figure 2. Height vs. time display of MPL normalized
backscattered energy observed on 21 September 1999.
model has a lower frequency of t < 0.2, and a higher
frequency of t larger than 0.3.
4. Cirrus Cloud Formation Mechanisms
[11] Both model and observations show that tropical
cirrus fall into two distinct categories: those observed near
convection (anvils) and those observed detached from
convection. Cirrus detached from convection are often thin,
laminar in appearance, and located near the tropical tropopause. Anvil cirrus are usually physically and optically
thicker than upper tropospheric (UT) cirrus located near
the TTL, have a lower cloud base height and have more
visible structure within the cloud. To assess the frequency
that the ECMWF model predicts each cirrus type, we
examine whether convective rain is present within a 1 hour
window of the cirrus event. Approximately 54% of cirrus
forms when no precipitation is present in the grid box over
the previous 1 hour.
[12] One possible mechanism for the formation and
maintenance of UT cirrus is the presence of large-scale
ascent. The model predicts vertical ascent in 77% of UT
cirrus cases when cloud base zb > 15 km (ascent is defined
as any value of w < 0 Pa/s at the same time that cirrus is
present in the grid-box). For cirrus clouds with zb > 7 km,
there is still a 75% frequency that cirrus occurs under
conditions of large-scale ascent.
[13] Tropical tropopause cirrus are also linked to cold
temperature perturbations in the upper troposphere. We
estimate that 76% of cases where all cloud layers in the
model column are located above 15 km are associated with
cold temperature anomalies, which is consistent with previous studies [Boehm and Verlinde, 2000; Comstock et al.,
2002]. To determine this, we use the 30 day temperature
anomaly located at the level of the model cloud top height.
Cold temperature perturbations in the tropical upper troposphere are sometimes linked with the occurrence of stratospheric waves [Boehm and Verlinde, 2000]. These results
indicate that the primary generation/maintenance mechanism for cirrus away from convection in the model is largescale ascent and cirrus are typically associated with cold
temperature anomalies.
L10106
and satellite measurements. As an example, we show
measurements and model output for a thin cirrus case
observed on 21 September 1999. For this typical UT cirrus
case, a thin cirrus cloud was observed by the MPL at Nauru
for nearly 12 hours (Figure 2). The depth of this cirrus
ranges from 200 m to 2 km and has an optical depth
ranging from 0.001 to 0.05. Small scale waves are apparent
in the lidar backscatter image, particularly between 1300
and 1800 UTC. Inspection of Geostationary Meteorological
Satellite (GMS) imagery (not shown) indicates convection
is not present in the region surrounding Nauru during the
cirrus event. The closest convection occurred to the south of
Nauru 4 – 5 days before the thin cirrus observation. This
implies that the source of the moisture for cirrus formation
could be convection but the formation mechanism is likely
another source, such as large-scale ascent.
[15] On this day, the ECMWF model predicts thin cirrus
above 15 km with a cloud cover varying between 50 and
100% (Figure 3a) and maximum ice water contents of less
than 1 mg kg 1 (Figure 3b). The cloud persists in a region
of large-scale ascent (Figure 3c). The model predicts no
precipitation in the grid box during the 24-hour period
around the cirrus event. The sloping layer observed by the
lidar may indicate sedimentation of ice crystals (Figure 2).
We estimate from Figure 2 that ice is falling at roughly
4 cm s 1. Therefore, the model assumption of zero fall
5. Analysis of a Thin Cirrus Case Study
[14] To analyze the ability of the ECMWF model to
predict specific cirrus cases observed over Nauru, we
examined numerous cases that exhibited characteristics
typical to anvil and UT cirrus clouds as judged by lidar
Figure 3. ECMWF model output for 21 September 1999.
Each 100 sub-boxes are averaged at 1 hour time intervals.
(a) Cloud cover, (b) IWC (g kg 1), and (c) vertical pressure
velocity (Pa s 1).
3 of 4
L10106
COMSTOCK AND JAKOB: TROPICAL CIRRUS CLOUD PROPERTIES
velocity for small ice crystals maybe incorrect. Although the
model did not simulate the sloping nature of the cloud, the
model does correctly predict the dissipation time to be
1800 UTC. The difference in cloud thickness is not
surprising given that the model vertical resolution at 15 km
is 1 km, whereas the lidar is 0.03 km. Also apparent in
the model vertical velocity are tropospheric waves just
below cloud base that appear to modulate the cirrus cloud
cover and IWC. During periods of peak downward motion
(red contours 0800 UTC) the cirrus cloud cover and IWC
decrease in magnitude. Thicker sections and higher IWC
occur when the downward motion decreases (0300 UTC).
This phenomena does not necessarily hold true at
1400 UTC; however, the magnitude of the downward
vertical velocity is not as strong.
[16] We found similar results after examination of several
cirrus cases observed over Nauru. Of the 32 cases examined, 56% were associated with convection, and 66% were
influenced by tropospheric waves (61% of anvils and 71%
of UT cirrus) as determined in the model vertical velocity
field. On occasion, the model failed to predict cirrus when it
was detected by the lidar (21% of UT cirrus cases). During
these times, the model predicted insufficient water vapor
and there was an absence of vertical upwelling in the upper
troposphere.
6. Summary and Future Work
[17] In this study, we have compared cloud properties
derived from the ECMWF model and measurements at the
ARM site located on Nauru Island. In addition to composite
statistics, as an example, we also examined a thin tropopause cirrus case to evaluate the ECMWF model’s ability to
forecast UT cirrus and under what conditions. Although
there is some difficulty comparing a model grid box with
single point measurements, our preliminary results show
that the model predicts both convective anvils and isolated
tropopause cirrus reasonably well. Cirrus persisting near the
tropopause under convectively suppressed conditions occur
when the model predicts overall ascent. Model vertical
velocity indicates that cirrus morphology is often modulated
by tropospheric waves.
[18] Although it seems obvious that tropopause cirrus
clouds form during periods of large-scale ascent, the mechanism that produces these conditions is not straightforward,
and there are few measurements to verify the large-scale
environment in the TTL. Several studies have attempted to
explain the formation and persistence of tropical UT cirrus
clouds using models and measurements. Some of the
current theories for UT cirrus formation are ice nucleating
by slow large-scale ascent or in shear turbulent mixing
[Jensen et al., 1996], or by cold temperature perturbations
in the UT caused by stratospheric waves [Boehm and
Verlinde, 2000]. One theory explains the presence of
large-scale ascent in the TTL as the result of momentum
transport by convectively driven Rossby waves in the ITCZ
region [Boehm and Lee, 2003]. Given the demonstrated
skill of the model to predict cirrus, studying the frequency
of these phenomena as well as identifying tropical cirrus
formation mechanisms and large-scale features using the
L10106
ECMWF model simulations combined with ARM measurements appears to be a promising avenue for future work.
[19] Acknowledgments. This research was supported by the DOE
Office of Biological and Environmental Research under contract numbers
DE-AC06-76RL01830 and LANL-23662-001-013T as part of the Atmospheric Radiation Measurement Program. The Pacific Northwest National
Laboratory is operated by Battelle for the U.S. Department of Energy
(DOE).
References
Ackerman, T. P., and G. M. Stokes (2003), The Atmospheric Radiation
Measurement program, Physics Today, 56, 38.
Boehm, M. T., and J. Verlinde (2000), Stratospheric influence on upper
tropospheric tropical cirrus, Geophys. Res. Lett., 27(19), 3209 – 3212.
Boehm, M. T., and S. Lee (2003), The implications of tropical Rossby
waves for tropical tropopause cirrus formation and for the equatorial
upwelling of the Brewer-Dobson circulation, J. Amos. Sci., 60, 247 – 261.
Campbell, J. R., D. L. Hlavka, E. J. Welton, C. J. Flynn, D. D. Turner, J. D.
Spinhirne, V. S. Scott III, and I. H. Hwang (2002), Full-time, eye-safe
cloud and aerosol lidar observation at atmospheric radiation measurement
program sites: Instruments and data processing, J. Atmos. Oceanic Technol., 19, 431 – 442.
Clothiaux, E. E., T. P. Ackerman, G. G. Mace, K. P. Moran, R. T.
Marchand, M. A. Miller, and B. E. Martner (2000), Objective determination of cloud heights and radar reflectivities using a combination of active
remote sensors at the ARM CART sites, J. Appl. Meteorol., 39, 645 – 665.
Comstock, J. M., and K. Sassen (2001), Retrieval of cirrus cloud radiative
and backscattering properties using combined lidar and infrared radiometer (LIRAD) measurements, J. Atmos. and Ocean. Tech., 18, 1658 –
1673.
Comstock, J. M., T. P. Ackerman, and G. G. Mace (2002), Ground-based
lidar and radar remote sensing of tropical cirrus clouds at Nauru Island:
Cloud statistics and radiative impacts, J. Geophys. Res., 107(D23), 4714,
doi:10.1029/2002JD002203.
Ebert, E. E., and J. A. Curry (1992), A parameterization of ice cloud optical
properties for climate models, J. Geophys. Res., 97(D4), 3831 – 3836.
Jakob, C., and S. A. Klein (1999), The role of vertically varying cloud
fraction in the parametrization of microphysical processes in the ECMWF
model, Q. J. R. Meteorol. Soc., 125, 941 – 965.
Jakob, C. (2001), The representation of cloud cover in atmospheric general
circulation models, PhD-thesis, Ludwig-Maximilians-Univ., Munich,
193 pp.
Jensen, E. J., O. B. Toon, H. B. Selkirk, J. D. Spinhirne, and M. R.
Schoeberl (1996), On the formation and persistence of subvisible cirrus
clouds near the tropical tropopause, J. Geophys. Res., 101(D16), 21,361 –
21,375.
Mace, G. G., C. Jakob, and K. P. Moran (1998), Validation of hydrometeor
occurrence predicted by the ECMWF model using millimeter wave radar
data, Geophys. Res. Lett., 25(10), 1645 – 1648.
McFarquhar, G. M., A. J. Heymsfield, J. Spinhirne, and B. Hart (2000),
Thin and subvisual tropopause tropical cirrus: Observations and radiative
impacts, J. Atmos. Sci., 57, 1841 – 1853.
Moran, K. P., B. E. Martner, M. J. Post, R. A. Krofli, D. C. Welsh, and K. B.
Widener (1998), An unattended cloud-profiling radar for use in climate
research, Bull. Am. Meteorol. Soc., 79, 443 – 455.
Rosenfield, J. E., D. B. Considine, M. R. Schoeberl, and E. V. Browell
(1998), The impact of subvisible cirrus clouds near the tropical tropopause on stratospheric water vapor, Geophys. Res. Lett., 25(11), 1883 –
1886.
Sherwood, S. C. (1999), On moistening of the tropical troposphere by cirrus
clouds, J. Geophys. Res., 104(D10), 11,949 – 11,960.
Tiedtke, M. (1993), Representation of clouds in large-scale models, Mon.
Wea. Rev., 121, 3030 – 3061.
Wielicki, B. A., T. M. Wong, R. P. Allan, A. Slingo, J. T. Kiehl, B. J. Soden,
C. T. Gordon, A. J. Miller, S. K. Yang, D. A. Randall, F. Robertson,
J. Susskind, and H. Jacobowitz (2002), Evidence for large decadal variability in the tropical mean radiative energy budget, Science, 295, 841 –
844.
J. M. Comstock, Pacific Northwest National Laboratory, P.O. Box 999
MSIN K9-24, Richland, WA 99352, USA. ( [email protected])
C. Jakob, Bureau of Meteorology Research Centre, GPO Box 1289K,
Melbourne, VIC 3001, Australia. ([email protected])
4 of 4